Companion for Trakt

Connect your own Trakt account to Claude or Codex for recommendations, viewing history, title search, lists, and release calendars. Version 3.0.0 replaces the generic API tools with curated tools that each perform one Trakt action, including requested changes to your lists, watchlist, history, ratings, and collection. Seven bundled skills guide these workflows; writes require explicit intent and a connection with write permission.

Desktop setup

Claude

1. Open Customize → Plugins → Add → Add marketplace → Add from a repository.

2. Enter swack-tools/trakt-ai-plugin and add the marketplace.

3. Install Companion for Trakt, open the plugin's Connectors tab, add the Trakt connector if needed, and choose Connect. Then start a new chat.

Codex

Add the marketplace once from a terminal:

codex plugin marketplace add swack-tools/trakt-ai-plugin

Restart the Codex app, open Plugins, choose the Companion for Trakt marketplace, and install Companion for Trakt. Start a new chat and follow the connection prompt.

Related MCP server: Trakt MCP Server

Command-line setup

Claude Code

claude plugin marketplace add swack-tools/trakt-ai-plugin
claude plugin install trakt-mcp@trakt-mcp
claude

In the new session, open /mcp, select the Trakt server, and authenticate.

Codex command-line tool

codex plugin marketplace add swack-tools/trakt-ai-plugin
codex plugin add trakt-mcp@trakt-mcp
codex

Follow the connection prompt when you first use Trakt.

Connect your Trakt account

Each person signs in with their own Trakt account:

1. On the authorization page, choose Connect Trakt.

2. Sign in to Trakt and approve access.

3. Choose Return to your app, then return to your chat.

Seven bundled workflows

• what-to-watch: a concise shortlist grounded in Trakt recommendations.

• watching-profile: retrieve every watched-summary page, or summarize the latest 100 watch events per medium.

• find-title: disambiguate titles with bounded search, or follow every page when you explicitly request all results.

• lists-and-watchlist: discover public lists, create personal lists, and make requested item changes.

• upcoming-releases: read personal or public calendars for a bounded date range and time zone.

• manage-library: record or remove specific watches, ratings, and collection items.

• connection-help: diagnose installation, OAuth, device codes, and rate limits.

Use natural language or your client’s skill picker. See the plugin README for setup, workflow details, and platform limits.

API coverage and write access

Thirty-seven MCP tools each perform one fixed Trakt action, with fixed inputs defined in api/trakt/tools.json. They accept no free-form path, method, or request body. Every write requires explicit confirmation and write authorization, list changes also verify ownership, and writes are paced to Trakt's limit of one per second. Successful calls include typed structuredContent while retaining the original JSON text for older clients. See the API reference.

New browser OAuth connections request read and write access when scope is omitted, and the consent page states those permissions. Explicit trakt:read requests stay read-only. Existing tokens and standalone device login remain read-only; reauthorize through your client with trakt:read trakt:write for writes. Refresh cannot increase scope. Each write tool also requires confirmed: true for a concrete change the user requested. Skills verify changes through reads and never automatically repeat a write after an uncertain failure.

Your data

Trakt tokens are stored encrypted with a key held only by your AI clients; records are deleted 30 days after last use or on sign-out. It is decrypted in memory only while handling a request. See the privacy page for sign-out and retention limits.

Try asking

• Recommend five movies for me using Trakt.

• What TV shows does Trakt recommend for me?

• Recommend science-fiction movies from 2020 onward.

• Show me the movies and TV shows I’ve watched.

• Search Trakt for Arrival.

• Create a private list called Rainy Sunday and add Arrival (2016).

• Show my upcoming episodes next week in America/Chicago.

• Rate Arrival (2016) 8.

Recommendations come from Trakt’s personalized results. By default, history returns watched-title summaries with play counts, one compact page per tool call. Compact detail keeps identifying metadata, genres, counts, and watch timestamps; request full detail only for additional metadata. The skills follow pagination.next_page to completion for all-history requests and recommendations based on all viewing. Ask for recently watched to use the latest 100 chronological watch events per medium, including repeat watches. Interrupted traversals are reported as partial with a resume page.

For standalone MCP or skills-only setup, see the installation guide. Skills alone still require the authenticated MCP connection.

Companion for Trakt is an independent, unofficial integration by SwackTech LLC. Companion for Trakt uses Trakt but is not endorsed or certified by Trakt. Repository installation does not establish approval by a public directory. Directory eligibility, publisher verification, and reviewer access remain owner responsibilities.

See the full documentation for other clients, troubleshooting, API details, and self-hosting.

Licensed under GPL-3.0-only.